Unverified50% confidenceFactExact time
The safety training data for models (human feedback in RLHF) is heavily concentrated in English, while multilingual capability is acquired from broader pretraining corpora, creating a systematic misalignment in coverage between the two.
1
Sources
50%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
Unverified不同大语言模型因训练数据分布、RLHF偏好对齐策略和模型架构差异，在同一问题上呈现风格迥异的回应倾向79% similarUnverified大语言模型的指令跟随能力是通过RLHF训练习得的统计倾向，注意力机制无法从架构层面区分硬规则与软建议78% similarVerified大语言模型在RLHF训练阶段，因人类标注员倾向于给更详细、更礼貌的回答打高分，导致模型学会了过度冗长的表达模式78% similarUnverifiedSome systems embed safety preferences into model weights during the training phase through RLHF (Reinforcement Learning from Human Feedback), causing the model itself to tend to refuse generating certain content rather than relying on external filtering.77% similarUnverified提示词与模型版本存在深度绑定关系，不同版本模型在RLHF训练数据、指令微调策略上的差异会导致相同提示词产生不同输出，这种现象被称为「提示词漂移」77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/933161API
curl https://kongchang.com/api/v1/knowledge/claims/933161MCP
get_claim(id=933161)